Evolutionary Algorithm Inspired by the Methods of Quantum Computer Sciences for the Improvement of a Neural Model of the Electric Power Exchange

Abstract

<jats:p>The work contains results of research on the possibility to improve the neural model of the Electric Power Exchange (polish: Towarowa Giełda Energii Elektrycznej – TGEE) in MATLAB and Simulink environment using evolutionary algorithm inspired by quantum computer science. The developed artificial neural network was trained using data for the Day Ahead Market, assuming the joint volume of supplied and sold electrical energy [MWh] as the input quantities in each hour of the 24-hour day, and average prices [PLN/MWh] as output quantities. The obtained model of the exchange system was improved using the evolutionary algorithm, and further improvement in the accuracy of the model by supplementing the evolutionary algorithm using quantum solutions, related to the initial population, crossover and mutation operators, selection, etc. were proposed.</jats:p>

Untitled

“…In order to implement the methods used in quantum computing, real data has been transformed into binary and quantum ones. The proposed method of quantizing and dequantizing real numbers into quantum numbers [21][22][23][24][25] use the principle of superposition, which for the qubit has the form:…”

“…This means that two states can be used to obtain the states of mixed value: first by drawing from the dominating range (numbers in the range <0.71 ÷ 1>) ket 0 or ket 1 respectively and the second method by drawing from recessive intervals (numbers in the range <0 ÷ 0.71>) ket 1 or ket 0. In the literature, the Hadamard (H) gate is most commonly used for quantizing mixed numbers, as a singleton quantum gate representing by a 2-dimensional unitary matrix, which is an alternative proposition of quantization given in [18,[23][24][25]:…”

“…The Hadamard gateway function for base vectors | 1> and |0> can be represented as follows: In order to determine the value of the activation function quantum matrix net 1 1 , dequantization operation was performed. In this case, an innovative solution using perceptron of ANN was proposed in which as input to the neural network the quantum mixed numbers described in the above square matrix expressing the quasi-mixed probability modules of pure states are implemented, and the real value as the output, the values obtained from the ANN (value net i j ) [21][22][23][24][25]. Selected results of comparative studies between the quantum neural network adder net i for the first neuronal output layer for the first input and analogue values of the perceptron adder of the ANN are shown in Figure 1.…”

“…At the same time artificial intelligence methods used in power system control are rapidly growing. They are frequently used in smart metering, modelling of on-line power transactions, in the development of micro-networks, and in numerous other areas of power engineering with respect to the various subsystems of the power system seen as a technical system, as well as broader -viewed as a technicaleconomic system [7,14,17,[21][22][23][24][25]29].…”

“…It is assumed, for example, networks that use ICT to reduce the costs and increase efficiency, and integrate distributed sources of electricity, including renewable energy. Therefore, the smart grid concept requires interdisciplinary research and application of modern technologies solutions to make system models more flexible, which requires a combined view of power engineering processes with ICT processes [21][22][23][24][25][26][27], by using quantum computing solutions.…”